Standard Performance Metrics Research Articles

Weakly supervised deep learning can analyze focal liver lesions in contrast-enhanced ultrasound.

Introduction Assessing the malignancy of focal liver lesions is an important yet challenging aspect of routine patient care. Contrast-enhanced ultrasound (CEUS) has proved to be a highly reliable tool but is very dependent on the examiner's expertise. The emergence of artificial intelligence has opened doors to algorithms that could potentially aid in the diagnostic process. In this study, we evaluate the performance of a weakly supervised deep learning model in classifying focal liver lesions (FLL) as malignant or benign. Methods Our retrospective feasibility study was based on a cohort of patients from a tertiary care hospital in Germany undergoing routine CEUS examination to evaluate malignancy of FLL. We trained a weakly supervised attention-based multiple instance learning algorithm during 5-fold cross-validation to distinguish malignant from benign liver tumors, without using any manual annotations, only case labels. We aggregated the on-average best performing cross-validation cycle and tested this combined model on a held-out test set. We evaluated its performance using standard performance metrics and developed explainability methods to gain insight into the model's decisions. Results We enrolled 370 patients, comprising a total of 955,938 images extracted from CEUS videos or manually captured during the examination. Our combined model was able to identify malignant lesions with a mean area under the receiver operating curve of 0.844 in the cross-validation experiment and 0.94 (95% CI 0.89 - 0.99) in the held-out test set. The accuracy, sensitivity, specificity, and F1-Score of the combined model in finding malignant lesions in the held-out test, yielded 80.0%, 81.8%, 84.6%, and 0.81, respectively. Our exploratory analysis using visual explainability methods revealed that the model appears to prioritize information that is also highly relevant to expert clinicians in this task. Conclusions Weakly supervised deep learning can classify malignancy in CEUS examinations of FLLs and thus might one day be able to assist doctors' decision-making in clinical routine.

Developing an interpretable machine learning model for diagnosing gout using clinical and ultrasound features.

Developing an interpretable machine learning model for diagnosing gout using clinical and ultrasound features.

Enhancing Multi-Disease Prediction with Machine Learning: A Comparative Analysis and Hyperparameter Optimization Approach

Although traditional methods based on statistical parameters are still important in healthcare, Machine learning (ML) algorithms offer promising results for analyzing health data. Therefore, the presented work aimed to evaluate the success of several supervised ML models with hyperparameter optimization (HPO) for predicting multiple diseases such as diabetes, heart disease, Parkinson's disease, and breast cancer. We evaluated seven distinct algorithms: Logistic Regression (LR), Gradient Boosting (GB), k-Nearest Neighbors (k-NN), Extreme Gradient Boosting (XGB), Support Vector Machines (SVM), Random Forests (RF), and a basic "nonlinear mapping technique". Each algorithm was trained and compared in isolation for each targeted health condition. The success of these techniques was assessed using standard performance metrics like accuracy, precision, F1-score, and recall. Additionally, hyperparameter optimization was applied to each algorithm and its effect on the result was observed. The results show the potential of ML for multiple disease prediction with individual models achieving high accuracy for specific diseases. SVM achieved 100% accuracy for heart disease, Gradient Boosting achieved 90% for diabetes, a simple Neural Network achieved 99% for breast cancer, and Random Forest achieved 100% for Parkinson's disease. These results emphasize the importance of selecting appropriate models for specific disease prediction tasks. A web-based application has been developed so that users can easily use the models by selecting a disease, providing relevant input, and receiving a prediction based on the chosen model. In conclusion, this study highlights the potential of machine learning and hyperparameter optimization for multi-disease prediction and underlines the importance of model selection.

Inception networks, data augmentation and transfer learning in EEG-based photosensitivity diagnosis

Abstract Photosensitivity refers to a neurophysiological condition in which the brain generates epileptic discharges known as Photoparoxysmal Responses (PPR) in response to light flashes. In severe cases, these PPR can lead to epileptic seizures. The standardized diagnostic procedure for this condition is called Intermittent Photic Stimulation. During this procedure, the patient is exposed to a flashing light, aiming to trigger these epileptic reactions while preventing their full development. Meanwhile, brain activity is monitored using Electroencephalography, which is visually analyzed by clinical staff to identify these responses. Hence, the automatic detection of PPR becomes a highly unbalanced problem that has been barely studied in the literature due to photosensitivity’s low prevalence. 
This research tackles this problem and proposes using Inception-based Deep Learning (DL) neural networks that, together with transfer learning, are trained in epilepsy seizure detection and tuned in the PPR automatic detection task. A Data Augmentation (DA) technique is also applied to balance the available data set, evaluating its effects on the DL models. The proposal outperformed state-of-the-art solutions in the literature, achieving higher ratios on standard performance metrics, and with DA significantly improving the Sensitivity without affecting Accuracy and Specificity. This project is currently being developed with patients from Burgos University Hospital, Spain.

Radiomics-based diagnosis of patellar chondromalacia using sagittal T2-weighted images.

This study aimed to explore and evaluate anovel method for diagnosing patellar chondromalacia using radiomic features from patellar sagittal T2-weighted images (T2WI). The experimental data included sagittal T2WI images of the patella from 40patients with patellar chondromalacia and 40healthy volunteers. The training set comprised 30cases of chondromalacia and 30healthy volunteers, while the test set included 10cases of each. Amachine learning algorithm was used to train the classification model, which was then evaluated using standard performance metrics. In the training set, the model achieved24 true negatives(TN), 18true positives(TP), 12false negatives(FN), and sixfalse positives(FP). Sensitivity, specificity, accuracy, and F1score for the training set were 0.6, 0.8, 0.7, and 0.667, respectively. The model achieved sixtrue negatives, eighttrue positives, twofalse negatives, and fourfalse positives in the test set. Sensitivity, specificity, accuracy, and F1score for the test set were 0.8, 0.6, 0.7, and 0.727, respectively. The radiomic analysis method based on patellar sagittal fat-suppressed T2WI images demonstrates good diagnostic capability for patellar bone marrow edema.

Comparison of Resampling Methods and Radiomic Machine Learning Classifiers for Predicting Bone Quality Using Dual-Energy X-Ray Absorptiometry.

Background/Objectives: This study presents a novel approach, based on a combination of radiomic feature extraction, data resampling techniques, and machine learning algorithms, for the detection of degraded bone structures in Dual X-ray Absorptiometry (DXA) images. This comprehensive approach, which addresses the critical aspects of the problem, distinguishes this work from previous studies, improving the performance achieved by the most similar studies. The primary aim is to provide clinicians with an accessible tool for quality bone assessment, which is currently limited. Methods: A dataset of 1531 spine DXA images was automatically segmented and labelled based on Trabecular Bone Score (TBS) values. Radiomic features were extracted using Pyradiomics, and various resampling techniques were employed to address class imbalance. Three machine learning classifiers (Logistic Regression, Support Vector Machine (SVM), and XGBoost) were trained and evaluated using standard performance metrics. Results: The SVM classifier outperformed the other classifiers. The highest F-score of 97.5% was achieved using the Grey Level Dependence Matrix and Grey Level Run Length Matrix feature combination with SMOTEENN resampling, which proved to be the most effective resampling technique, while the undersampling method yielded the lowest performance. Conclusions: This research demonstrates the potential of radiomic texture features, resampling techniques, and machine learning methods for classifying DXA images into healthy or degraded bone structures, which potentially leads to improved clinical diagnosis and treatment.

Precision in Parsing: Evaluation of an Open-Source Named Entity Recognizer (NER) in Veterinary Oncology.

Integrating Artificial Intelligence (AI) through Natural Language Processing (NLP) can improve veterinary medical oncology clinical record analytics. Named Entity Recognition (NER), a critical component of NLP, can facilitate efficient data extraction and automated labelling for research and clinical decision-making. This study assesses the efficacy of the Bio-Epidemiology-NER (BioEN), an open-source NER developed using human epidemiological and medical data, on veterinary medical oncology records. The NER's performance was compared with manual annotations by a veterinary medical oncologist and a veterinary intern. Evaluation metrics included Jaccard similarity, intra-rater reliability, ROUGE scores, and standard NER performance metrics (precision, recall, F1-score). Results indicate poor direct translatability to veterinary medical oncology record text and room for improvement in the NER's performance, with precision, recall, and F1-score suggesting a marginally better alignment with the oncologist than the intern. While challenges remain, these insights contribute to the ongoing development of AI tools tailored for veterinary healthcare and highlight the need for veterinary-specific models.

Spatial Performance Indicators for Traffic Flow Prediction

Traffic flow prediction, crucial for traffic management, relies on spatial and temporal data to achieve high accuracy. However, standard performance metrics only measure the average prediction errors and overlook the spatiotemporal aspects. To address this gap, this study introduces three simple spatial key performance indicators (KPIs): Global Moran’s I, Getis-Ord General G, and Adapted PageRank Algorithm Modified (APAM). We evaluated the traffic prediction results for synthetic clustering scenarios and four different prediction methods applied to the PeMSD8 dataset using spatial KPIs. Spatial KPIs are calculated based on traffic prediction errors and the adjacency matrix of the traffic network. Our results demonstrate that spatial KPIs can effectively differentiate between synthetic clustering scenarios. Global Moran’s I measures the spatial autocorrelation, Getis-Ord General G measures the spatial clustering of high/low values, and the univariate analysis of APAM deduces the distribution of node importance by considering node centrality and node values. Getis-Ord General G showed the highest sensitivity, being capable of distinguishing between methods with similar average RMSE, whereas Global Moran’s I and APAM univariate analysis revealed subtle differences between methods. Spatial KPIs serve as important complementary metrics for performance evaluation in the design of robust traffic management systems.

Deep FS: A Deep Learning Approach for Surface Solar Radiation.

Contemporary environmental challenges are increasingly significant. The primary cause is the drastic changes in climates. The prediction of solar radiation is a crucial aspect of solar energy applications and meteorological forecasting. The amount of solar radiation reaching Earth's surface (Global Horizontal Irradiance, GHI) varies with atmospheric conditions, geographical location, and temporal factors. This paper presents a novel methodology for estimating surface sun exposure using advanced deep learning techniques. The proposed method is tested and validated using the data obtained from NASA's Goddard Earth Sciences Data and Information Services Centre (GES DISC) named the SORCE (Solar Radiation and Climate Experiment) dataset. For analyzing and predicting accurate data, features are extracted using a deep learning method, Deep-FS. The method extracted and provided the selected features that are most appropriate for predicting the surface exposure. Time series analysis was conducted using Convolutional Neural Networks (CNNs), with results demonstrating superior performance compared to traditional methodologies across standard performance metrics. The proposed Deep-FS model is validated and compared with the traditional approaches and models through the standard performance metrics. The experimental results concluded that the proposed model outperforms the traditional models.

Deep insights into gastrointestinal health: A comprehensive analysis of GastroVision dataset using convolutional neural networks and explainable AI

The gastrointestinal (GI) tract is critical in digestion and nutrient absorption, thus vital for human health. However, it is prone to diseases like cancer. Manual assessments introduce accuracy variations, consistency issues, and delays. Resources like the GastroVision dataset were introduced to advance AI in this field. Yet, it faces class imbalance issues, and baseline evaluation lacks novel methodologies, impacting accuracy. We propose a novel deep-learning model to enhance accuracy and robustness. Our approach involves averaging weights of multiple models fine-tuned with diverse hyper-parameters. In contrast to classical ensembles, our approach uses DenseNet-121 as a baseline and enables the averaging of numerous models without incurring extra inference or memory costs. Data augmentation techniques are incorporated to address class imbalance. We achieve promising results on standard performance metrics, substantially improving over baseline, notably 2.4% points in Macro Precision. Additionally, we integrate explainable AI (XAI) techniques to enhance reliability and interpretability, shedding light on the model’s decision-making processes. Our study contributes to robust methodologies for imbalanced datasets, promoting model transparency and trust in predictive outcomes for clinical decision support systems.

Recent developments in microfluidic passive separation to enable purification of platelets for transfusion.

Platelet transfusion is a lifesaving therapy intended to prevent and treat bleeding. However, in addition to platelets, a typical unit also contains a large volume of supernatant that accumulates multiple pro-inflammatory contaminants, including residual leukocytes, microaggregates, microparticles, antibodies, and cytokines. Infusion of this supernatant is responsible for virtually all adverse reactions to platelet transfusions. Conventional methods for removing residual leukocytes (leukoreduction) and reducing the volume of transfused supernatant (volume reduction) struggle to mitigate these risks holistically. Leukoreduction filters can remove leukocytes and microaggregates but fail to reduce supernatant volume, whereas centrifugation can reduce volume, but it is ineffective against larger contaminants and damages platelets. Additionally, platelet purification based on these methods is often too logistically complex, time-consuming, and labor-intensive to implement routinely. Emerging microfluidic technologies offer promising alternatives through passive separation mechanisms that enable cell separation with minimal damage and drastically reduced instrumentation size and facility requirements. This review examines recent innovations in microfluidic cell separation that can be used for leukoreduction and volume reduction of platelets. It begins by defining the performance requirements that any separation method must meet to successfully replace conventional methods currently used to perform these tasks. Standard performance metrics are described, including leukocyte depletion efficiency, degree of volume reduction, processing throughput, and platelet recovery. Finally, the review outlines the primary challenges that must be overcome to enable simple-to-use, disposable microfluidic devices capable of both reducing the platelet unit volume and removing pro-inflammatory contaminants, while preserving most functional platelets for transfusion.

Examining the Influenza A Virus Sialic Acid Binding Preference Predictions of a Sequence-Based Convolutional Neural Network.

Though receptor binding specificity is well established as a contributor to host tropism and spillover potential of influenza A viruses, determining receptor binding preference of a specific virus still requires expensive and time-consuming laboratory analyses. In this study, we pilot a machine learning approach for prediction of binding preference. We trained a convolutional neural network to predict the α2,6-linked sialic acid preference of influenza A viruses given the hemagglutinin amino acid sequence. The model was evaluated with an independent test dataset to assess the standard performance metrics, the impact of missing data in the test sequences, and the prediction performance on novel subtypes. Further, features found to be important to the generation of predictions were tested via targeted mutagenesis of H9 and H16 proteins expressed on pseudoviruses. The final model developed in this study produced predictions on a test dataset correctly 94% of the time and an area under the receiver operating characteristic curve of 0.93. The model tolerated about 10% missing test data without compromising accurate prediction performance. Predictions on novel subtypes revealed that the model can extrapolate feature relationships between subtypes when generating binding predictions. Finally, evaluation of the features important for model predictions helped identify positions that alter the sialic acid conformation preference of hemagglutinin proteins in practice. Ultimately, our results provide support to this in silico approach to hemagglutinin receptor binding preference prediction. This work emphasizes the need for ongoing research efforts to produce tools that may aid future pandemic risk assessment.

Dandelion center detection in perennial ryegrass with Heat maps using Convolutional Neural Networks

Dandelion center detection in perennial ryegrass with Heat maps using Convolutional Neural Networks

A Novel Method for 3D Lung Tumor Reconstruction Using Generative Models.

Lung cancer remains a significant health concern, and the effectiveness of early detection significantly enhances patient survival rates. Identifying lung tumors with high precision is a challenge due to the complex nature of tumor structures and the surrounding lung tissues. To address these hurdles, this paper presents an innovative three-step approach that leverages Generative Adversarial Networks (GAN), Long Short-Term Memory (LSTM), and VGG16 algorithms for the accurate reconstruction of three-dimensional (3D) lung tumor images. The first challenge we address is the accurate segmentation of lung tissues from CT images, a task complicated by the overwhelming presence of non-lung pixels, which can lead to classifier imbalance. Our solution employs a GAN model trained with a reinforcement learning (RL)-based algorithm to mitigate this imbalance and enhance segmentation accuracy. The second challenge involves precisely detecting tumors within the segmented lung regions. We introduce a second GAN model with a novel loss function that significantly improves tumor detection accuracy. Following successful segmentation and tumor detection, the VGG16 algorithm is utilized for feature extraction, preparing the data for the final 3D reconstruction. These features are then processed through an LSTM network and converted into a format suitable for the reconstructive GAN. This GAN, equipped with dilated convolution layers in its discriminator, captures extensive contextual information, enabling the accurate reconstruction of the tumor's 3D structure. The effectiveness of our method is demonstrated through rigorous evaluation against established techniques using the LIDC-IDRI dataset and standard performance metrics, showcasing its superior performance and potential for enhancing early lung cancer detection. This study highlights the benefits of combining GANs, LSTM, and VGG16 into a unified framework. This approach significantly improves the accuracy of detecting and reconstructing lung tumors, promising to enhance diagnostic methods and patient results in lung cancer treatment.

Feature-level fusion of face and speech based multimodal biometric attendance system with liveness detection

The rapid growth of deep learning and the Internet of Things has spurred the need for touchless biometric systems in areas where cleanliness and non-intrusive user interaction are critical. In order to authenticate seamlessly, traditional biometric methods such as fingerprint, hand, etc. recognition require physical contact and, therefore, risk hygiene issues, hence making face and speaker verification more viable alternatives. A robust Multimodal Biometric Attendance System (MBAS) is needed due to the vulnerabilities and limitations of single modality systems. In this research, we introduce MBAS using feature-level fusion of speech data with face data, combining the best of both worlds. The textural features based on a person’s facial appearance are integrated with dynamic speech information for liveness detection, followed by dimensionality reduction using linear discriminant analysis, and then incorporated into a Bi-LSTM classifier. Therefore, for better security, accuracy, and anti-spoofing attacks, this approach is proposed in addition to increasing accuracy as well as enhancing security against spoofing attacks. Two publicly available datasets, DeepfakeTIMIT and AVSpeech, are extensively explored to evaluate different fusion strategies, classifier types, and standard performance metrics. The proposed system outperformed other cutting-edge biometric based systems by exhibiting a 97.51% high accuracy rate with a precision of 99.10% and an equal error rate of 2.48%. These findings affirm the effectiveness and possible real-world applications of the MBAS concept, along with its enhancement ensuring safety. Furthermore, this study underscores the importance of incorporating advanced liveness detection into secure contactless biometrics solutions for modern attendance management in various industries that encompass both face and voice modalities.

Improvement The Accuracy of Convolutional Neural Network with Using Undersampling Method on Unbalanced Credit Card Dataset

In this study, we address the challenge of imbalanced data in credit card fraud detection by proposing a novel approach that leverages Convolutional Neural Networks (CNNs) and undersampling techniques. The imbalance in the dataset, typical of real-world financial transactions, often leads to biased models favoring the majority class. To mitigate this, we employ undersampling to balance the classes, thereby enhancing the CNN's ability to learn from minority instances crucial for fraud detection. Our method is validated on a large unbalanced credit card dataset, demonstrating significant improvements in accuracy compared to traditional CNN models trained on imbalanced data. We evaluate our approach using standard performance metrics, including precision, recall, and F1-score, showcasing its effectiveness in accurately identifying fraudulent transactions while minimizing false positives. Furthermore, we pro-vide insights into the CNN's decision-making process through visualization techniques, shedding light on its ability to discern fraudulent patterns within the data. Our findings highlight the importance of addressing class imbalance in fraud detection tasks and underscore the efficacy of undersampling in enhancing the performance of deep learning models, particularly CNNs, in handling imbalanced datasets.

A Linear Time Shrinking-SL(t)-ViT Approach for Brain Tumor Identification and Categorization

Automated brain tumor detection and classification systems have gained popularity in recent years because traditional diagnosis procedures are time-consuming and costly in nature. Deep learning(DL) methods, specifically pre-trained convolutional neural networks (CNNs), have shown promising results in accurately and rapidly classifying brain tumors. However, the lack of diverse magnetic resonance imaging (MRI) datasets has hindered the ability of DL algorithms to generalize effectively. To address this issue, this paper proposes a DL model using generative adversarial networks (GANs) in conjunction with data augmentation strategies and a structural similarity loss function employed for generating annotated images. A novel DL model inspired from Vision Transformer, Shrinking Linear Time Vision Transformer (SL(t)-ViT) network model is proposed for disease data classification. The model underwent extensive evaluation across multiple datasets, employing standard performance metrics to assess its efficacy in brain tumor identification. The proposed model achieved remarkable testing accuracies of 0.995, 0.996, 0.9954, 0.998, and 0.997 for binary classification tasks across multiple datasets, and 0.986, 0.982, 0.985, and 0.993 for multi-class classification tasks. These results underscore the superior performance of our proposed model, showcasing its capability to outperform state-of-the-art techniques. Specifically, it demonstrated a substantial margin of improvement, ranging from 1-2% for binary classification and 9-10% for multi-class classification, solidifying its position as a leading approach in brain tumor identification. Results demonstrate the efficacy of the proposed model, outperforming other models. Overall, this study highlights the potential of SL(t)-ViT and GANs in improving the accuracy, resource consumption, and efficiency of brain tumor diagnosis.

Spatial attention guided cGAN for improved salient object detection

Recent research shows that Conditional Generative Adversarial Networks (cGANs) are effective for Salient Object Detection (SOD), a challenging computer vision task that mimics the way human vision focuses on important parts of an image. However, implementing cGANs for this task has presented several complexities, including instability during training with skip connections, weak generators, and difficulty in capturing context information for challenging images. These challenges are particularly evident when dealing with input images containing small salient objects against complex backgrounds, underscoring the need for careful design and tuning of cGANs to ensure accurate segmentation and detection of salient objects. To address these issues, we propose an innovative method for SOD using a cGAN framework. Our method utilizes encoder-decoder framework as the generator component for cGAN, enhancing the feature extraction process and facilitating accurate segmentation of the salient objects. We incorporate Wasserstein-1 distance within the cGAN training process to improve the accuracy of finding the salient objects and stabilize the training process. Additionally, our enhanced model efficiently captures intricate saliency cues by leveraging the spatial attention gate with global average pooling and regularization. The introduction of global average pooling layers in the encoder and decoder paths enhances the network's global perception and fine-grained detail capture, while the channel attention mechanism, facilitated by dense layers, dynamically modulates feature maps to amplify saliency cues. The generated saliency maps are evaluated by the discriminator for authenticity and gives feedback to enhance the generator's ability to generate high-resolution saliency maps. By iteratively training the discriminator and generator networks, the model achieves improved results in finding the salient object. We trained and validated our model using large-scale benchmark datasets commonly used for salient object detection, namely DUTS, ECSSD, and DUT-OMRON. Our approach was evaluated using standard performance metrics on these datasets. Precision, recall, MAE and Fβ score metrics are used to evaluate performance. Our method achieved the lowest MAE values: 0.0292 on the ECSSD dataset, 0.033 on the DUTS-TE dataset, and 0.0439 on the challenging and complex DUT-OMRON dataset, compared to other state-of-the-art methods. Our proposed method demonstrates significant improvements in salient object detection, highlighting its potential benefits for real-life applications.

An analytical approach for power system frequency stability evaluation

AbstractRecent years have seen high penetration of renewable energies, which have significantly reduced the inertia of bulk power systems. As a result, the frequency behaviour of power systems is becoming more complex. To resolve this technical challenge, there is a particularly strong interest in developing analytical solutions for frequency dynamics studies. This study first describes a second‐order frequency dynamics model for power systems with renewable energies. A non‐linear perturbation approach is suggested to drive the analytical solution of the model. It is shown that, under many circumstances, frequency dynamics can be effectively approximated using a linear model. Subsequently, the article describes a fourth‐order linear frequency dynamics model that takes into account governor‐turbines. A polynomial eigenvalue method is proposed to identify the dominant and non‐dominant modes of the solution of the four‐order model. It is demonstrated that the dominant mode has a decisive impact on frequency behaviour, while the non‐dominant modes influence the relative frequency oscillations only. Finally, the study derives the analytical expressions of the standard frequency performance metrics and examines the impact of damping and inertia parameters. The introduced results are verified using two test systems, demonstrating the accuracy and effectiveness of the suggested method.

A framework for classifying breast cancer via heterogenetic attention mechanism and optimized feature selection

Breast cancer poses a significant threat to women’s health, emphasizing the crucial role of timely detection. Traditional pathology reports, though widely used, face challenges prompting the development of automated Deep Learning (DL) tools. DL models, gaining traction in radiology, offer precise diagnoses; however, issues with generalization on varying dataset sizes persist. This paper introduces a computationally efficient DL framework, addressing dataset imbalance through a hybrid model design, ensuring both accuracy and speed in breast cancer image classification. Proposed model novel design excels in accuracy and generalization across medical imaging datasets, providing a robust tool for precise diagnostics. The proposed model integrates features from two classifiers, Inception ResNet V2 and Vision Transformers (ViT), to enhance the classification of breast cancer. This synergistic blend enhances adaptability, ensuring consistent performance across diverse dataset scales. A key contribution is the introduction of an Efficient Attention Mechanism within one of the classifiers, optimizing focus on critical features for improved accuracy and computational efficiency. Further, a Resource-Efficient Optimization model through feature selection is proposed, streamlining computational usage without compromising accuracy. Addressing the inherent heterogeneity within classifiers, our framework integrates high dimensional features comprehensively, leading to more accurate tumor class predictions. This consideration of heterogeneity marks a significant leap forward in precision for breast cancer diagnosis. An extensive analysis on datasets, BreakHis and BACH, that are imbalanced in nature is conducted by evaluating complexity, performance, and resource usage. Comprehensive evaluation using the datasets and standard performance metrics accuracy, precision, Recall, F1-score, MCC reveals the model’s high efficacy, achieving a testing accuracy of 0.9936 and 0.994, with precision, recall, F1-score and MCC scores of 0.9919, 0.987, 0.9898, 0.9852 and 0.989, 1.0, 0.993, 0.988 on the BreakHis and BACH datasets, respectively. Our proposed model outperforms state-of-the-art techniques, demonstrating superior accuracy across different datasets, with improvements ranging from 0.25% to 15% on the BACH dataset and from 0.36% to 15.02% on the BreakHis dataset. Our results position the framework as a promising solution for advancing breast cancer prediction in both clinical and research applications. The collective contributions, from framework and hybrid model design to feature selection and classifier heterogeneity consideration, establish a holistic and state-of-the-art approach, significantly improving accuracy and establishing optimization in breast cancer classification from MRI images. Future research for the DL framework in breast cancer image classification includes enhancing interpretability, integrating multi-modal data, and developing personalized treatments.